# Grok independent review — EM-DAT-07

Source conversation: https://grok.com/c/7ec0c4c2-a63b-48b3-b868-e123f11de303

## Verdict

Accept with conditions. WM-ACT-036 masters Analytical Study; WM-ACT-034 supplies reusable assessment mechanics and is not the master. Recommendation may specialize WM-KNW-007 as a deontic claim if claim mood distinguishes assertoric findings from deontic recommendations and WM-KNW-010, WM-ACT-024 and WM-REC-002 remain distinct. Analysis Method remains identifier-unassigned.

## Evidence and counterexample

The proposed scope, version pins, assumptions, defeaters, uncertainty and causal gates are inquiry rules and therefore belong to the Research Study boundary rather than to a scoring event. Findings and recommendations share proposition structure while differing in mood.

A one-shot pass/fail vendor-control assessment uses WM-ACT-034 without a sampling frame, reusable method version or competing causal readings. Calling it an Analytical Study would over-apply the type. Conversely, treating a published recommendation as authorized would erase the separate decision and approval authorities.

## Identity and method boundary

WM-ACT-036 masters the study. WM-ACT-034 may be used for assessment steps but is not an alternate parent. A reusable Analysis Method is distinct from Study, Finding, Recommendation, Assessment and Report. It owns a named procedure, version, applicability, required inputs and permitted output claim kinds. A change to applicability or output kind creates a new method revision; issued findings are never rewritten.

## Causality and recommendation boundary

A causal finding is a WM-KNW-007 claim and is incomplete without measurand, comparator, design and competing-explanation analysis. Observation is not cause. Alternative interpretations remain distinct conflict-linked claims. Sampling bias, a missing comparator and unversioned data are explicit defeaters. Ranking does not eliminate a rival claim; elimination requires an explicit defeater.

A recommendation specializes WM-KNW-007 as an ought-claim. It has no authority bit. Proposed, superseded and withdrawn are claim-lifecycle states rather than approval. Acceptance is a WM-ACT-024 activity that cites a WM-KNW-010 rationale and the selected recommendation. WM-REC-002 may present findings and recommendations but does not own them.

## Scenario

Ticket count in an opt-in support channel rises over an interval. The study pins ticket extract, routing configuration, staffing roster and method versions and records selection bias. A descriptive finding records the increase with uncertainty. Competing causal findings for productivity decline and routing change each require their own measurand, comparator and design and remain conflict-linked. A recommendation to freeze routing remains non-authoritative until a separate decision rationale and authorized occurrence exist.

## Invariants

1. Analytical Study profiles WM-ACT-036 and is not mastered by WM-ACT-034.
2. Use of WM-ACT-034 does not type an activity as Analytical Study.
3. Analysis Method has no identifier here and remains distinct from Study, Finding, Recommendation and Report.
4. A Finding pins study, method, sample and data versions.
5. Studies may share a method version; findings may not silently share unpinned method or data versions.
6. A causal finding requires measurand, comparator, design and a competing-explanation set.
7. Alternative interpretations remain distinct conflict-linked WM-KNW-007 claims.
8. Undischarged sampling-frame bias blocks causal closure.
9. Recommendation is a deontic claim and carries no authorization or selected-rationale identity.
10. An authorized decision requires both WM-KNW-010 and WM-ACT-024; WM-REC-002 substitutes for neither.
11. Claim mood is mandatory and a finding or recommendation cannot exceed the study scope.
12. Changed study, method, sample or data versions create a successor finding rather than an overwrite.

## Minimum completion shape and blockers

Study records identity, question, scope, design, sample frame, data pins, method pin, assumptions, defeaters and uncertainty policy. Method records named procedure, version, applicability, required inputs and allowed claim kinds. Finding records assertoric mood, pins, class, uncertainty, assumptions and defeaters, plus the causal gate where applicable. Recommendation records deontic mood, cited findings, target and conditions without approval fields.

Analysis Method remains identifier-unassigned. A registered conflict-link vocabulary, a versioning object model and a required mood discriminator are still absent. Existing workflows may resist the strict causal-completeness gate; this is an adoption hold, not grounds to weaken the boundary.